Blob Blame History Raw
From: Alex Ng <alexng@messages.microsoft.com>
Date: Sun, 6 Aug 2017 13:12:53 -0700
Patch-mainline: v4.14-rc1
Subject: Drivers: hv: balloon: Correctly update onlined page count
Git-commit: 6df8d9aaf3afe25aacf20c69022a4c6d57b77a95
References: fate#323887

Previously, num_pages_onlined was updated using value from memory online
notifier. This is incorrect because they assume that all hot-added pages
are online, even though we only online the amount that's backed by the
host. We should update num_pages_onlined only when the balloon driver
marks a page as online.

Signed-off-by: Alex Ng <alexng@messages.microsoft.com>
Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: Olaf Hering <ohering@suse.de>
---
 drivers/hv/hv_balloon.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/hv/hv_balloon.c b/drivers/hv/hv_balloon.c
--- a/drivers/hv/hv_balloon.c
+++ b/drivers/hv/hv_balloon.c
@@ -584,10 +584,6 @@ static int hv_memory_notifier(struct notifier_block *nb, unsigned long val,
 
 	switch (val) {
 	case MEM_ONLINE:
-		spin_lock_irqsave(&dm_device.ha_lock, flags);
-		dm_device.num_pages_onlined += mem->nr_pages;
-		spin_unlock_irqrestore(&dm_device.ha_lock, flags);
-		/* Fall through */
 	case MEM_CANCEL_ONLINE:
 		if (dm_device.ha_waiting) {
 			dm_device.ha_waiting = false;
@@ -644,6 +640,9 @@ static void hv_page_online_one(struct hv_hotadd_state *has, struct page *pg)
 	__online_page_set_limits(pg);
 	__online_page_increment_counters(pg);
 	__online_page_free(pg);
+
+	WARN_ON_ONCE(!spin_is_locked(&dm_device.ha_lock));
+	dm_device.num_pages_onlined++;
 }
 
 static void hv_bring_pgs_online(struct hv_hotadd_state *has,